We’ve been running a lot of Hugging Face model inference in our k8s clusters for some internal tools, and I’m starting to think about supply chain security for these models. We already use Wiz for cloud security posture, and I know they can scan container images in the CI/CD pipeline.
Has anyone integrated Wiz to scan Hugging Face model files (like `.bin`, `.safetensors`, or even the entire repo snapshot) as part of a CI/CD pipeline before deployment? I'm specifically curious about:
* The scanning mechanics – do you pull the model into a container and scan it, or does Wiz have a direct integration for model registries?
* What kinds of issues does it catch? Malware in serialized files, vulnerable dependencies in associated code, misconfigurations?
* Any performance impact on pipeline runtimes? Some of these models are several GB.
I’m picturing something in a GitLab CI step or GitHub Action that fails the build if a critical vulnerability is found. Right now we’re only scanning the runtime containers, but the model itself feels like a blind spot.
If you’ve set this up, I’d love to see a snippet of how you trigger the scan. Bonus points if you’re using the Wiz CLI or API in a multi-stage pipeline.
- away